GPUs can now use PCIe-attached memory or SSDs to boost VRAM capacity

Companies like AMD, Intel, and Nvidia may soon support Panmnesia's CXL IP, enabling GPUs to expand memory capacity using PCIe-attached memory or SSDs. Panmnesia's CXL IP offers low-latency memory expansion solutions for GPUs, addressing the increasing memory requirements for AI training datasets. By developing a CXL 3.1-compliant root complex and host bridge, Panmnesia allows GPUs to access external memory over PCIe, improving performance compared to traditional methods like UVM. Testing shows that Panmnesia's solution achieves significantly lower latency and faster execution times, outperforming UVM and CXL-Proto. While CXL support holds promise for AI/HPC GPUs, the adoption by major players like AMD and Nvidia remains uncertain. Whether these companies will integrate CXL support or develop their own technology in response to the trend of using PCIe-attached memory for GPUs is yet to be seen. Panmnesia's innovative approach demonstrates potential benefits for GPU memory expansion and performance optimization in the evolving landscape of AI and HPC applications.